NEW YORK — Researchers have developed a machine learning tool to predict which esophageal squamous cell carcinoma (ESCC) patients may respond to platinum-based neoadjuvant chemotherapy (NAC).
While that is the primary treatment for ESCC, the treatment response rate is between 57 percent and 72 percent. To look for biomarkers to predict sensitivity and response to NAC treatment, researchers from the Riken Center for Integrative Medical Sciences and elsewhere conducted immunogenetic analyses of ESCC samples collected from more than 120 patients before they underwent treatment. As they reported in Cell Reports Medicine on Monday, they then homed in on different features associated with a patient's later response or non-response to treatment. By bundling these features — which include indicators of neutrophil infiltration and certain copy number alterations — into a software tool, they could predict NAC response.
"[W]e identified genetic features and differences in immune reactivity that are uniquely associated with response to NAC. This indicates the presence of a subset of patients with pre-existing mutations that confer resistance to NAC," senior author Hidewaki Nakagawa from Riken and colleagues wrote in their paper. "Importantly, these mutations may be clinically valid and can support targeted treatment strategies that have been successful in patients with metastatic ESCC, as various agents are in clinical trials."
The researchers collected biopsy samples from 121 ESCC patients who then received NAC. Of these patients, eight had a complete response and 67 a partial response, while 36 had stable disease and 10 progressive disease.
While the researchers did not uncover significant differences in gene expression between patients who responded to treatment versus those who did not, they did detect gene pathways that were enriched among responders. These pathways included a number that are involved in the immune response, such as the IL2 STAT5 signaling and interferon gamma response pathways.
The team further examined T-cell signatures in the samples, as tumor-infiltrating CD8+ T cells have been able to predict immunotherapy or chemotherapy response and patient survival in other cancer types, and found a significant difference in treatment response based on these signatures. This led them to examine other immune cells in the tumor microenvironments, and they found that samples with high levels of CD8+ T cells, CD4+ T cells, and B cells had high response rates, while those with mostly neutrophils had low response rates.
At the same time, the researchers also examined the genomic profiles of the tumors. They noted similar mutational burdens between the tumors of responders and non-responders, and that nearly 80 percent of the ESCCs had copy number variants at the chromosome-arm level. Among responders, a number of recurrent copy number changes clustered in certain biological pathways, including the interferon gamma response pathway, suggesting that such alterations are associated with sensitivity to chemotherapy and immune response in ESCC. They further generated six copy number signatures found within ESCC that are associated with treatment response.
The researchers combined all these transcriptomic, immune, and CNV factors, alongside other indicators like smoking status, into a random forest method-based model to predict treatment response. After training it on a subset of their cohort and testing it on another, the researchers found that the model had more than 84 percent accuracy, and a sensitivity and specificity of 66.7 percent. In a cohort of 20 new cases, the model had an accuracy of 84 percent.
This suggested to the researchers that the factors they uncovered could be applied in the clinic to guide treatment strategies. "We envision a treatment prediction model that combines copy number changes and immune profiling of ESCC with clinical data to improve the response to NAC and patient prognosis," they wrote.